|Publication number||US6349113 B1|
|Application number||US 09/177,686|
|Publication date||Feb 19, 2002|
|Filing date||Oct 22, 1998|
|Priority date||Nov 3, 1997|
|Also published as||EP0913794A2, EP0913794A3|
|Publication number||09177686, 177686, US 6349113 B1, US 6349113B1, US-B1-6349113, US6349113 B1, US6349113B1|
|Inventors||Roland Mech, Joern Ostermann, Jurgen Stauder|
|Original Assignee||At&T Corp.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (8), Non-Patent Citations (7), Referenced by (55), Classifications (8), Legal Events (4)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This application claims the benefit of U.S. Provisional Application No. 60/064,107, filed Nov. 3, 1997, the entire contents of which are incorporated herein by reference.
The present invention relates generally to the processing of video image sequences. More specifically, the present invention relates to the detection of moving cast shadows within video image sequences so that object segmentation can be performed.
Shadows occur in a wide variety of scenes, including a wide variety of scenes in sequential video images. If shadows are identified in video images, they can provide a great deal of information about the shape, relative position, and surface characteristics of objects in a scene. Although humans can easily distinguish shadows from objects, identifying shadows by computers is more difficult. For example, identifying shadows by computers involves object segmentation, the separation of an object, its shadow and the background which can have stationary portions and moving portions.
Identifying shadows by computers is even more difficult with cast shadows that are moving from one video image to the next video image. Known methods and systems have been developed to attempt to detect moving cast shadows. For example, one known method performs shadow detection with a static camera providing the image by subdividing an image into blocks and computing the luminance contrast for each block; blocks are identified as moving cast shadows where the luminance contrast changes from pixel to pixel within a block. See Skiftstad, K. and Jain, R., “Illumination Independent Change Detection for Real World Image Sequences,” Computer Vision, Graphics, and Image Processing 46, 387-99 (1989).
This known method, however, suffers from shortcomings. Specifically, it cannot distinguish a moving cast shadow from a moving object where the object lacks texture, i.e., lacks sharp contrast within the object.
An image region changed by a moving cast shadow from a first image to a second image is detected. For each pixel within a change detection mask (a binary mask indicating image areas of difference between the first image and the second image) and a set of neighboring pixels, the following steps are performed. Whether the pixel and the set of neighboring pixels include a static background edge or no edge is determined. Whether the pixel and the set of neighboring pixels include an edge with a spatial signal step width greater than a threshold is determined. Whether the pixel and the set of neighboring pixels have a uniform temporal change of illumination is determined. The pixel is classified as being changed by a moving cast shadow when at least two of the above-mentioned determinations succeed for the pixel or when at least one determination from the above-mentioned determinations succeeds for a majority of the set of neighboring pixels.
In another embodiment of the present invention, detected image regions are used to estimate the two-dimensional shape of moving objects in image sequences even in the presence of moving cast shadows. In another embodiment, detected image regions are temporarily integrated to represent the total of moving cast shadows.
FIG. 1 shows a possible configuration for the formation of a cast shadow, according to an embodiment of the present invention.
FIG. 2 illustrates a process by which static background edges can be detected, according to an embodiment of the present invention.
FIG. 3 illustrates a process for testing the spatial constancy of the frame ratio within an image by evaluating its local spatial variance, according to an embodiment of the present invention.
FIG. 4 illustrates a process by which penumbras can be detected, according to an embodiment of the present invention.
FIG. 5 illustrates a luminance step model for a luminance step in an image perpendicular to a shadow contour, according to an embodiment of the present invention.
FIG. 6 illustrates the application of heuristic rules to determine image regions changed by a moving cast shadow, according to an embodiment of the present invention.
FIGS. 7 and 8 illustrate a process by which two-dimensional shape estimation of moving objects in an image sequence can be applied to image regions changed by moving cast shadows, according to an embodiment of the present invention.
FIG. 1 shows a possible configuration for the formation of a cast shadow, according to an embodiment of the present invention. Object 100 is illuminated by a light source 110. Cast shadow 130 is projected onto background 120. The cast shadow 130 includes shadow center 131 and shadow penumbras 132. Shadow penumbras 132 are soft transitions from dark to bright where some of the light from light source 110 reaches background 120. The appearance of cast shadow 130 can be recorded by a video camera 140.
Cast shadows 130 on background 120, including cast shadows which are moving (referred to herein as “moving cast shadows”), can be detected by the video camera 140 which collects consecutive images and by a processor (not shown) which analyzes the video images. The methods described below can be used by the processor to analyze the video images. Note that the processor need not be directly coupled to video camera 140; rather, the video images can be detected by the video camera 140 and later analyzed by the processor or the video images can be analyzed in near real time by the processor as each image is being detected by the video camera 140. In one implementation, the video images are detected by video camera 140, analyzed by the processor, and sent for transmission over a telecommunications network.
The appearance of a cast shadow in an image of a video camera 140 can be described by an image signal model. This model describes the image luminance, Sk, as follows:
where, k is the time instance; x,y is the two-dimensional image position; Ek (x,y) is the irradiance of the surface of object 100; and tk (x,y) is the reflectance of the surface of object 100. The irradiance Ek (x,y) is the amount of light power per receiving object surface area. The irradiance Ek (x,y) is a function of the direction, L, of the light source, the intensities cp and ca of light source 110 and ambient light, respectively, and the object surface normal, N, according to the following equation:
In equation 2, the term k(x,y), which has a value between 0 and 1, describes the transition inside penumbra 132 and depends on the scene geometry. The intensity cp of the light source is proportional to 1/r2 with r being the distance between object 100 and light source 110.
In the image signal model of equation 1, photometric distortions of perspective projection are neglected. Note also the gamma nonlinearity (i.e., the exponential power law used to approximate the curve of output magnitude versus the input magnitude over the region of interest) of a video camera is not considered. Finally, the color of the ambient light is assumed to be the same as the color of a point light source.
The following assumptions are made: the video camera 140 and background 120 are static; background 120 is locally planar and the position of light source 110 is distant from background 120; light source 110 has an extended shape as compared to the distance of the moving object 100; and the intensity, cp, of light source 110 is high. Consequently, moving cast shadows 130 on background 120 will be part of a change detection mask. A change detection mask indicates those image regions having a large frame difference between the previous and current image. In other words, the change detection mask indicates the difference in luminance on a per-pixel basis between two consecutive image frames.
To illustrate a consequence of the assumption that the intensity cp of light source 110 is high, consider a pixel at position x,y showing a part of background 120. Because the reflectance of background 120 is static and does not change, tk(x,y)=tk+1 (x,y) holds. When the pixel is in shadow center 131 at time k and in shadow penumbra 132 at time k+1, the difference in image luminance between frames will be large as illustrated by the following equation:
All pixels that are part of the change detection mask are evaluated by the following three criteria: (1) detection of static background edges, (2) detection of uniform changes of illumination and (3) penumbra detection. The results of these three criteria can be combined to a binary mask for regions of background 120 that are changed by moving cast shadows 130. Finally, for each image of an image sequence, moving cast shadows can be detected by temporal integration of regions changed by moving cast shadows. Each of the aspects will be discussed below.
1. Detection of Static Background Edges
To satisfy the assumption that video camera 140 and background 120 are static, a number of configurations are possible. First, any motion of video camera 140 and background 120 can be restricted. Alternatively, the previous image as Sk can be motion compensated with respect to the subsequent image Sk+1.
FIG. 2 illustrates a process by which static background edges can be detected according to an embodiment of the present invention. For example, in textured background regions inside the change detection mask, moving cast shadows 130 can be distinguished from moving objects 100 by their static background edges, because neither background 120 nor camera 140 are moving, or their motion is compensated.
At step 200, a first set of edges in a first image is detected. At step 210, a second set of edges in a second image is detected. At step 220, the first image is subtracted from the second image to produce the frame difference.
At step 230, a high-frequency filter is applied to the frame difference to produce a high frequency frame difference. At step 240, the high frequency frame difference is compared to a high-frequency threshold. The threshold for high-frequency activity can be adaptively calculated from the high frequency activities of the frame difference outside the change detection mask.
For each edge from the first and second set of edges from the first and second images, respectively, steps 250 and 260 are performed. At conditional step 250, a determination is made for each pixel of each edge whether the high frequency frame difference is below the threshold. Where high frequency frame difference is below the threshold, that edge is classified as a static edge in step 260.
The static edges can be used to detect moving cast shadows on a non-moving background inside the changed detection mask. As equation 1 illustrates, static edges in an image sk(x,y) can be rendered either to the reflectance t(x,y) or to the irradiance Ek (x,y). Static edges caused by discontinuities in the reflectance suggest a texture in the static background 120. Static edges caused by discontinuities in the irradiance suggest discontinuous shading at three-dimensional shape edges of a static background 120. Thus, static edges inside the change detection mask suggest the possibility of a moving cast shadow 130 on static background 120.
2. Detection of Uniform Changes of Lumination
Note the following assumptions: first, background 120 is assumed to be locally planar and light source 110 is assumed to be distant from background 120. Consequently, the irradiance as calculated by equation 2 is spatially constant because the surface normal N(x,y) is spatially constant. Further assume that the penumbra regions 132 of cash shadow 130 can be neglected.
FIG. 3 illustrates a process for testing the spatial constancy of the frame ratio within an image by evaluating its local spatial variance, according to an embodiment of the present invention. Steps 300 through 330 are performed for each pixel for two consecutive frame images. At step 300, the frame ratio is determined for each pixel within the change detection mask. The frame ratio for each pixel can be calculated using the following equation:
Because the luminance at position x, y for the pixel is assumed to be changed due to a moving cast shadow on a static background, the background reflectance can also be assumed to be unchanged and therefore tk (x,y)=t+1(x,y). Thus, neglecting any camera noise, the frame ratio can be simplified to
Consequently, where the luminance of position x, y is changed due to a moving cast shadow, the frame ratio is spatially constant in the neighborhood of x,y, because the irradiance is assumed to be constant as discussed above. Thus, if the frame ratio is locally spatially constant, a moving cast shadow can be assumed to position x, y. The frame ratio is then tested for spatial constancy by evaluating its local spatial variance.
At step 310, the local spatial variance of the frame ratio for the pixel is compared with an illumination threshold. A small variance in the local spatial variance should be allowed to address noise compensation.
At conditional step 320, the determination is made as to whether the local spatial variance is below the illumination threshold. If the local spatial variance is below the illumination threshold, then the process proceeds to conditional step 325. At conditional step 325, a determination is made as to whether the frame ratio in a local neighborhood (e.g., a 3×3 pixel area around the pixel being considered) is uniformly above or below one. If this condition is satisfied, then the process proceeds to step 330. At step 330, the pixel is classified as having a uniform temporal change of illumination. The occurrence of a pixel having a uniform temporal change of illumination suggests the presence of a moving cast shadow at that pixel location.
The illumination threshold can be adaptively calculated from the local variances of the frame ratio outside the change detection mask.
Note the process described in FIG. 3 erroneously can detect a moving cast shadow if at position x, y a uniformly colored moving object that rotates is visible. In this case, the simplification from equation 4 to equation 5 would still hold true and the frame ratio will still be locally, spatially constant although a moving cast shadow would not exist at this particular pixel location.
3. Penumbra Detection
Because the extent of light source 110 is assumed to be not negligible compared to the distance between light source 110 and moving object 100, cast shadow 130 has penumbra regions 132. The moving cast shadow 130 can be detected by the existence and characteristics of penumbras 132.
The penumbra 132 of cast shadow 130 causes a soft luminance step at the contour of a shadow 130. The luminance step in an image perpendicular to a shadow contour can be modeled by a luminance step model as illustrated in FIG. 5, according to an embodiment of the present invention. In this luminance step model, the luminance is assumed to rise linearly across shadow penumbra 132 from a low luminance inside a shadow (i.e., center shadow 131) to a high luminance outside the shadow (i.e., background 120). The luminance step within penumbra 132 can be characterized by its step height h, step width w and its gradient g which equals h/w If the width of a luminance step caused by a penumbra 132 is much larger than that of edges caused by the aperture of video camera 140 for surface texture edges of object 100 or edges of object 100, then the luminance step can be used for shadow detection.
Table 1 characterizes the luminance step height, gradient, and width of different kinds of edges in an image. As Table 1 shows, shadow edges can be distinguished from other edges by their luminance step width. The luminance step height alone is not an appropriate criterion by which shadow edges can be distinguished because either a shadow edge caused by a bright light source 100 or a texture edge with much contrast may cause a high luminance step. The luminance step gradient alone also is not an appropriate criterion by which shadow edges can be distinguished either because the gradient of a shadow edge caused by a bright light source 100 (with a certain extent) may be comparable to that of a texture edge with less contrast (and a small aperture for video camera 140).
shadow contour, bright light source
shadow contour, weak light source
texture/object edge, much contrast
texture/object edge, less contrast
FIG. 4 illustrates a process by which penumbras can be detected, according to an embodiment of the present invention. At step 400, moving cast shadow border candidates are obtained from the border of the change detection mask. The moving cast shadow border candidate can belong to edges of object 100 or a shadow contour, because the change detection mask contains image regions changed by moving objects 100 or moving cast shadows 130.
The number of moving cast shadow border candidates is low compared to the number of edges indicated by known edge detection algorithms. Moreover, known edge detection algorithms have difficulties in finding soft shadow contours. Additionally, the number of moving cast shadow border candidates is further reduced because the object mask of the first image, if available, is or-connected to the change detection mask to fill holes inside the change detection mask. Also, to enhance the precision, the moving cast shadow border candidates are moved perpendicular to the border of the change detection mask to a position of highest luminance gradient. The gradient is measured perpendicular to the border of the change detection mask using a Sobel operator. The Sobel operator consists of two finite-duration impulse response (FIR) filters with the filter
Steps 410, 420 and 430 are performed for each moving cast shadow border candidate. At step 410, a spatial signal step width of the frame difference is evaluated. At conditional step 420, the spatial signal step width is compared to a width threshold. If the spatial signal step width exceeds the width threshold, then the moving cast shadow border candidate is classified as a moving cast shadow border in step 430.
To determine the spatial signal step width for each moving cast shadow border candidate, the height and gradient of the signal steps perpendicular to the edge are measured for each moving cast shadow border candidate. The signal step height and gradient are measured in the frame difference between two consecutive images because whether the relevant edges are in the previous or in the current image depends on the unknown motion of cast shadows and objects. The signal step height can be measured by the difference of the average frame differences from both sides of the edge. For example, a 3 pixel by 3 pixel averaging window (for example, for the common intermediate format (CIF) image format) can be placed beside the edge. The signal gradient can be measured using a Sobel operator aligned perpendicular to the edge. The direction of the edge can be measured by a regression line evaluating moving cast shadow border candidates in a neighborhood of a 7 pixel by 7 pixel area.
The spatial signal step width, w, equals the height, h, divided by the gradient, g. The width threshold can be selected for the particular system as appropriate. For example, the width threshold can equal 2.4 pixels for a standard video camera. The width threshold can be lower for high definition television (HDTV) or higher for low light level video systems.
4. Detection of Image Regions Changed by Moving Cast Shadows
To detect image regions changed by moving cast shadows, the results of the three criteria from sections 1, 2 and 3 can be evaluated by heuristic rules. For each pixel of the change detection mask, a determination must be made whether the changes are caused by a moving cast shadow or by some other phenomena. Table 2 summarizes some of the heuristic evaluation rules to determine whether or not the pixel has been changed by a moving cast shadow. As the table illustrates, the first column considers whether a change was detected. The second column evaluates the result of edge classification from section 1. The third column indicates the result of illumination change classification from section 2 and the fourth column indicates the decision as to whether the pixel has been changed by a moving cast shadow. Additionally, the penumbra criteria from section 3 is evaluated in a local neighborhood of each pixel. If too many object edges are observed, the shadow hypothesis from Table 2, i.e, column 4, is rejected.
Decision: Pixel has changed
(see section 1)
by a moving cast shadow
if any neighboring pixel
satisfies row 13 then YES;
if any neighboring pixel
satisfies row 10 but not row
15, or rows 0-6 but not rows
12 and 17 then YES;
if any neighboring pixel
satisfies row 13 but not 11 and
17 then YES;
if any neighboring pixel
satisfies row 13 then YES;
FIG. 6 illustrates the application of heuristic rules to determine image regions changed by a moving cast shadow, according to an embodiment of the present invention. Steps 600 through 650 are repeated for each pixel within the change detection mask.
At step 600, a determination is made as to whether the pixel and the neighboring pixels include a static background edge. The determination performed in step 600 is made in accordance with the process described in section 1 above.
At step 610, a determination is made as to whether the pixel and the neighboring pixels are close to an edge with a spatial signal step width greater than a width threshold. The determination of step 610 can be made according to the process described in section 2 above.
At step 620, a determination is made as to whether the pixel and the neighboring pixels have a uniform temporal change of illumination. The determination made in step 620 can be performed according to the process described in section 3 above.
Steps 630 through 650 classify pixels as being changed by moving cast shadows according to the heuristic rules discussed in Table 2 and Section 3. At conditional step 630, the determinations made in step 600, 610 and 620 are evaluated with respect to the pixel and its neighboring pixels. If, however, at least two determinations for the pixel and its neighboring pixels do not succeed, then the process ends for that pixel. If at least two determinations succeed for the pixel and its neighboring pixel, then the process proceeds to step 650. At step 650, the pixel is classified as being changed by a moving cast shadow.
5. Segmentation of Moving Objects Considering Moving Shadows
FIGS. 7 and 8 illustrate a process by which a two-dimensional shape estimation of moving objects in an image sequence can be applied to image sequences containing both moving objects and moving cast shadows according to an embodiment of the present invention. At step 700, the apparent motion of video camera 140 or background 120 can be estimated and compensated to reflect any kind of global motion, e.g., caused by zoom and pan of the camera 140. At step 710, scene cut detection is performed by evaluating whether the mean square error between a current original frame Sk+1 and a camera motion compensated from the previous frame Sk, CMC exceeds a given threshold. If the threshold is exceeded, then all parameters are reset to their initial values. The scene cut detection performed in step 710 is only performed in the background regions of the previous frame which are taken from the previous object mask (OMk). In that mask, all pixels are set to foreground which belong to a moving object in the previous frame.
At step 720, a change detection mask between two successive frames is estimated. This step 720 is described in further detail with reference to FIG. 8 and will be discussed below.
At step 730, an initial object mask OMi is calculated by eliminating the uncovered background areas from the final change detection mask CDMk+1. Therefore, displacement information for pixels within the changed regions is used. The displacement is estimated by a hierarchical blockmatcher (HBM). See M. Bierling, “Displacement estimation by hierarchical block matching”, 3rd SPIE Symposium on Visual Communications and Image Processing, Cambridge, U.S.A., pp. 941-51, Nov. 1988, which is incorporated herein by reference for background. For a higher accuracy of the calculated displacement vector field, the change detection mask from the first step is considered by the HBM. Uncovered background is detected by pixels with foot- or top-point of the corresponding displacement vector being outside the changed area in the final change detection mask, CDMk+1.
At step 740, the final object mask is estimated. In other words, the boundaries of the initial object mask, OMiare adapted to luminance edges in the current image to improve the accuracy. The final result is the final object mask OMk+1.
FIG. 8 describes the process by which the change detection mask is estimated in step 720 of FIG. 7, according to an embodiment of the present invention. At step 800, an initial change detection mask, CDMi, is determined based on a first image and a second image. In other words, an initial change detection mask, CDMi, between the two successive frames is generated by thresholding the frame difference using a global threshold. In this initial change detection mask, pixels with image luminance change due to a moving object are labeled as changed, others are labeled as unchanged.
At step 810, a shadow portion of the change detection mask changed by the moving cast shadow is detected to produce a remaining portion of the change detection mask. The process of detecting the shadow portion of a change detection mask which is changed by the moving cast shadow can be done according to the processes described above with reference to sections 1, 2 and 3.
At step 820, boundaries of the changed image areas within the remaining portion of the change detection mask are smoothed. These boundaries can be smoothed, for example, by a relaxation technique using locally adapted thresholds. Consequently, the process adapts framewise automatically to noise from video camera 140.
At step 830, the smoothed remaining portion of the change detection mask is combined with an object mask from a first image, if available, to produce an object change detection mask. This step allows the production of temporally stable object regions. More specifically, the object change detection mask contains all pixels from the remaining portion of a change detection mask which are labeled as changed, and additionally, all pixels which belong to the object mask of the previous frame. This is based on the assumption that all pixels which belong to the previous object mask should belong to the current object change detection mask. Moreover, to avoid infinite error propagation, a pixel from the previous object mask is only labeled as changed in the object change detection mask if it was also labeled as changed in the remaining portion of the change detection mask of one of the last N frames. The value N corresponds to the time period which this particular pixel has been identified as being changed. The value N adapts automatically to the sequence by evaluating the size and motion amplitudes of the moving objects in the previous frame.
At step 840, small regions from the object change detection mask are eliminated resolving in the final change detection mask CDMk+1.
It should, of course, be understood that while the present invention has been described in reference to a particular system configuration and process, other systems and processes should be apparent to those of ordinary skill in the art. For example, althought the present invention was described in reference to an example of an arrangement of the object, light source, background and video camera, other arrangements are possible.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5481628||Mar 9, 1993||Jan 2, 1996||Eastman Kodak Company||Method and apparatus for spatially variant filtering|
|US5557684 *||Dec 27, 1994||Sep 17, 1996||Massachusetts Institute Of Technology||System for encoding image data into multiple layers representing regions of coherent motion and associated motion parameters|
|US5631976 *||Apr 29, 1994||May 20, 1997||International Business Machines Corporation||Object imaging system|
|US5832115 *||Jan 2, 1997||Nov 3, 1998||Lucent Technologies Inc.||Ternary image templates for improved semantic compression|
|US5940538 *||Aug 5, 1996||Aug 17, 1999||Spiegel; Ehud||Apparatus and methods for object border tracking|
|US6026183 *||Oct 25, 1996||Feb 15, 2000||Texas Instruments Incorporated||Content-based video compression|
|US6069918 *||Oct 26, 1996||May 30, 2000||Robert Bosch Gmbh||Method of detecting moving objects in chronologically successive images|
|US6075875 *||Sep 30, 1996||Jun 13, 2000||Microsoft Corporation||Segmentation of image features using hierarchical analysis of multi-valued image data and weighted averaging of segmentation results|
|1||Hyung-Il Choi et al., "Motion Interpretation by Analyzing Difference Images", Journal of Systems Architecture, Elsevier Science B.V., Mar. 1, 1997, pp. 307-316.|
|2||Illumination Independent Change Detection for Real World Image Sequences, K. Skifstad and R. Jain, Computer Vision, Graphics, and Image Processing, pp. 387-399, vol. 46, No. 3, Jun. 1989.|
|3||Image Segmentation Based on Object Oriented Mapping Parameter Estimation, M. Hötter and R. Thoma, Signal Processing, pp. 315-334, vol. 15, No. 3, Oct. 1988.|
|4||Image Segmentation Based on Object Oriented Mapping Parameter Estimation, Michael Hötter and Robert Thomas, Signal Processing 15 (1988), p. 315-334.|
|5||Marciej Orkisz, "Moving Objects Locations in Complex Scenes Filmed by a Fixed Camera", vol. 9, No. 4, Sep. 1992, pp. 325-346.|
|6||P. Stelmaszyk et al, "Dynamic Scenes Analysis by Moving Edges Detection", 1985, pp. 1181-1193.|
|7||Shadow Segmentation and Classification in a Constrained Environment C. Jiang and M. Ward, Image Understanding, pp. 213-225, vol. 59, No. 2, Mar. 1994.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US6539055 *||Dec 3, 1999||Mar 25, 2003||Intel Corporation||Scene change detector for video data|
|US6625310 *||Mar 23, 2001||Sep 23, 2003||Diamondback Vision, Inc.||Video segmentation using statistical pixel modeling|
|US6819778 *||Mar 29, 2001||Nov 16, 2004||Nec Corporation||Method and system for tracking a fast moving object|
|US6845178 *||Jun 27, 2001||Jan 18, 2005||Electro Scientific Industries, Inc.||Automatic separation of subject pixels using segmentation based on multiple planes of measurement data|
|US6904159 *||Dec 20, 2001||Jun 7, 2005||Mitsubishi Electric Research Laboratories, Inc.||Identifying moving objects in a video using volume growing and change detection masks|
|US7039107 *||Sep 1, 1999||May 2, 2006||Siemens Aktiengesellschaft||Method and array for processing a digitized image with image points|
|US7106898 *||Dec 6, 2000||Sep 12, 2006||California Institute Of Technology||3D scanning using shadows|
|US7224852||Sep 22, 2003||May 29, 2007||Objectvideo, Inc.||Video segmentation using statistical pixel modeling|
|US7313254 *||Sep 27, 2002||Dec 25, 2007||Lg Electronics Inc.||Method for extracting object region|
|US7366323 *||Feb 19, 2004||Apr 29, 2008||Research Foundation Of State University Of New York||Hierarchical static shadow detection method|
|US7423645 *||Jun 1, 2005||Sep 9, 2008||Microsoft Corporation||System for softening images in screen space|
|US7424175||Feb 27, 2007||Sep 9, 2008||Objectvideo, Inc.||Video segmentation using statistical pixel modeling|
|US7639878||Nov 17, 2005||Dec 29, 2009||Honeywell International Inc.||Shadow detection in images|
|US7672530||Jan 27, 2006||Mar 2, 2010||Tandent Vision Science, Inc.||Method and system for identifying illumination flux in an image|
|US7784107 *||Jun 1, 2007||Aug 24, 2010||Victor B. Kley||High speed measurement, analysis and imaging systems and methods for length scales from meter to sub-nanometer|
|US7826640 *||Apr 29, 2008||Nov 2, 2010||State University New York||Hierarchical static shadow detection method|
|US7873219||Mar 23, 2009||Jan 18, 2011||Tandent Vision Science, Inc.||Differentiation of illumination and reflection boundaries|
|US7970168 *||Oct 25, 2010||Jun 28, 2011||The Research Foundation Of State University Of New York||Hierarchical static shadow detection method|
|US8045761 *||May 30, 2006||Oct 25, 2011||Intelliview Technologies Inc.||Detection of environmental conditions in a sequence of images|
|US8159499 *||May 8, 2008||Apr 17, 2012||Disney Enterprises, Inc.||Rendering of shadows with hand-painted appearance|
|US8265417 *||Oct 16, 2008||Sep 11, 2012||Sony Corporation||Image processing apparatus, method, and program for adding shadow information to images|
|US8306261 *||Jun 13, 2008||Nov 6, 2012||International Business Machines Corporation||Detection of an object in an image|
|US8457401||Mar 1, 2007||Jun 4, 2013||Objectvideo, Inc.||Video segmentation using statistical pixel modeling|
|US8564661||Jul 26, 2007||Oct 22, 2013||Objectvideo, Inc.||Video analytic rule detection system and method|
|US8577170 *||Sep 15, 2011||Nov 5, 2013||Microsoft Corporation||Shadow detection in a single image|
|US8711217||Dec 15, 2005||Apr 29, 2014||Objectvideo, Inc.||Video surveillance system employing video primitives|
|US8761497 *||Dec 10, 2009||Jun 24, 2014||Rafael Advanced Defense Systems Ltd.||Removal of shadows from images in a video signal|
|US8776261||Aug 4, 2009||Jul 8, 2014||Victor B. Kley||Tool tips with scanning probe microscopy and/or atomic force microscopy applications|
|US8861788 *||Jul 3, 2012||Oct 14, 2014||International Business Machines Corporation||Detection of an object in an image|
|US9020261||May 3, 2013||Apr 28, 2015||Avigilon Fortress Corporation||Video segmentation using statistical pixel modeling|
|US20010046309 *||Mar 29, 2001||Nov 29, 2001||Toshio Kamei||Method and system for tracking a fast moving object|
|US20020024593 *||Dec 6, 2000||Feb 28, 2002||Jean-Yves Bouguet||3D scanning using shadows|
|US20030063669 *||Sep 27, 2002||Apr 3, 2003||Lee Jin Soo||Method for extracting object region|
|US20030118214 *||Dec 20, 2001||Jun 26, 2003||Porikli Fatih M.||Identifying moving objects in a video using volume growing and change detection masks|
|US20040151374 *||Sep 22, 2003||Aug 5, 2004||Lipton Alan J.||Video segmentation using statistical pixel modeling|
|US20050162515 *||Feb 15, 2005||Jul 28, 2005||Objectvideo, Inc.||Video surveillance system|
|US20060177137 *||Jan 27, 2006||Aug 10, 2006||Tandent Vision Science, Inc.||Differentiation of illumination and reflection boundaries|
|US20060177149 *||Jan 27, 2006||Aug 10, 2006||Tandent Vision Science, Inc.||Method and system for identifying illumination flux in an image|
|US20060274064 *||Jun 1, 2005||Dec 7, 2006||Microsoft Corporation||System for softening images in screen space|
|US20070110309 *||Nov 17, 2005||May 17, 2007||Honeywell International Inc.||Shadow detection in images|
|US20070280504 *||May 30, 2006||Dec 6, 2007||Wael Badawy||Detection of environmental conditions in a sequence of images|
|US20080095435 *||Dec 21, 2006||Apr 24, 2008||Objectvideo, Inc.||Video segmentation using statistical pixel modeling|
|US20080100704 *||Dec 15, 2005||May 1, 2008||Objectvideo, Inc.||Video surveillance system employing video primitives|
|US20080135750 *||Jun 1, 2007||Jun 12, 2008||Kley Victor B||High Speed Measurement, Analysis and Imaging Systems and Methods For Length Scales From Meter to Sub-Nanometer|
|US20090033673 *||May 8, 2008||Feb 5, 2009||Disney Enterprises, Inc.||Rendering of shadows with hand-painted appearance|
|US20090103831 *||Oct 16, 2008||Apr 23, 2009||Yusuke Nakamura||Image processing apparatus, image processing method, and program therefor|
|US20090297023 *||Mar 1, 2007||Dec 3, 2009||Objectvideo Inc.||Video segmentation using statistical pixel modeling|
|US20090310821 *||Dec 17, 2009||Connell Ii Jonathan H||Detection of an object in an image|
|US20100026802 *||Jul 26, 2007||Feb 4, 2010||Object Video, Inc.||Video analytic rule detection system and method|
|US20100031405 *||Aug 4, 2009||Feb 4, 2010||Kley Victor B||Tool Tips with Scanning Probe Microscopy and/or Atomic Force Microscopy Applications|
|US20100074539 *||Mar 25, 2010||Tandent Vision Science, Inc.||Differentiation of illumination and reflection boundaries|
|US20110273620 *||Dec 10, 2009||Nov 10, 2011||Rafael Advanced Defense Systems Ltd.||Removal of shadows from images in a video signal|
|US20120269397 *||Oct 25, 2012||International Business Machines Corporation||Detection of an object in an image|
|WO2009018050A1 *||Jul 23, 2008||Feb 5, 2009||Disney Entpr Inc||Rendering of shadows with hand-painted appearance|
|WO2010073172A2 *||Dec 10, 2009||Jul 1, 2010||Rafael Advanced Defense Systems Ltd.||Removal of shadows from images in a video signal|
|U.S. Classification||375/240.08, 348/700, 348/699|
|Cooperative Classification||G06K9/346, G06T7/2053|
|European Classification||G06T7/20D, G06K9/34R|
|Oct 22, 1998||AS||Assignment|
Owner name: AT&T CORP., NEW YORK
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:OSTERMANN, JOERN;REEL/FRAME:009537/0883
Effective date: 19981021
|Jun 30, 2005||FPAY||Fee payment|
Year of fee payment: 4
|Jun 22, 2009||FPAY||Fee payment|
Year of fee payment: 8
|Mar 18, 2013||FPAY||Fee payment|
Year of fee payment: 12